Hello,Your paper is truly excellent! Reading through it, I found the logic very clear and smooth. However, the only issue is that I noticed you mentioned needing 16 A100 GPUs, which is indeed a bit challenging. My current setup can only support 8 RTX 3090 Ti GPUs. I was wondering if it would be possible for me to reproduce your work under these constraints? I saw in your response to another issue that it might be feasible to first obtain the teacher model's outputs and then perform distillation. Could you kindly provide some guidance on how to approach this process?
I also have a small question to ask. I am an undergraduate student just starting to learn about artificial intelligence. My advisor has assigned me the task of using the POPE metric to evaluate model hallucination. I noticed that the GitHub project includes clear evaluation scripts. Am I correct in assuming that my main tasks would be:
1、Providing a model for evaluation
2、Preparing a dataset that meets the requirements outlined in the POPE paper
I am eagerly looking forward to your response. Thank you so much for your time and help!
Hello,Your paper is truly excellent! Reading through it, I found the logic very clear and smooth. However, the only issue is that I noticed you mentioned needing 16 A100 GPUs, which is indeed a bit challenging. My current setup can only support 8 RTX 3090 Ti GPUs. I was wondering if it would be possible for me to reproduce your work under these constraints? I saw in your response to another issue that it might be feasible to first obtain the teacher model's outputs and then perform distillation. Could you kindly provide some guidance on how to approach this process?
I also have a small question to ask. I am an undergraduate student just starting to learn about artificial intelligence. My advisor has assigned me the task of using the POPE metric to evaluate model hallucination. I noticed that the GitHub project includes clear evaluation scripts. Am I correct in assuming that my main tasks would be:
1、Providing a model for evaluation 2、Preparing a dataset that meets the requirements outlined in the POPE paper I am eagerly looking forward to your response. Thank you so much for your time and help!
学长您好！您的这篇论文写的真的太好了，读起来感觉逻辑很流畅。但是惟一的问题是看见您标注需要16块A100……这个确实是有点困难，我目前的设备只够支持8卡3090Ti，不知道是否可以允许我复现您的工作？我看您回答的另一个Issue时有说可以先将教师模型的输出获得后再进行蒸馏，请问这个过程能否大致指点一下？
另有个小问题想请教您，我是刚刚接触人工智能的本科生，导师目前布置的任务是用POPE指标衡量模型的幻觉，我看到Github项目里有很清晰的评估脚本，那我需要做的是不是就是： 1、提供模型 2、对照POPE论文提供符合其论文要求的数据集
十分期待您的回复，万分感激！